Predictive Probabilistic and Predictive Possibilistic Models for Risk Assess- ment in Grid Computing

We show a hybrid probabilistic and possibilistic model for assessing the risk of a service level agreement for a computing task in a cluster/grid environment. Using the predictive probabilistic approach we develop a framework for resource management in grid computing, and by introducing an upper limit for the number of failures we approximate the probability that a particular computing task is successful. In the possibilistic model we estimate the degree of possibility of success using a fuzzy nonparametric regression technique using the possibility distribution for the maximal number of failures derived from a resource provider’s observations. The probabilistic models scale from 10 nodes to 100 nodes and then on to any (reasonable) number of nodes; in the same fashion also the possibilistic models scale to 100 nodes and then on to any (reasonable) number of nodes. The resource provider can use both the probabilistic and the possibilistic models to get two alternative risk assessments and then he can choose either the probabilistic or the possibilistic risk assessment; or the hybrid model – which is a cautious/conservative approach.